TinyML Datasets as Enablers of 6G Edge Intelligence: Key Insights and Research Gaps

Tiny Machine Learning (TinyML) enables intelligent inference on resource-constrained edge devices, positioning it as a cornerstone for sixth-generation (6G) wireless systems demanding ubiquitous, low-latency, distributed intelligence at unprecedented scale. However, TinyML development is significantly constrained by limited availability of representative datasets and absence of standardized benchmarking protocols aligned with 6G requirements. This survey analyzes 12 representative TinyML datasets across five critical application domains, selected for 6G relevance, edge compatibility, and validation potential. Experimental results demonstrate practical feasibility: 25.9<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><mml:math display="inline"><mml:mo>×</mml:mo></mml:math></inline-formula> compression with <0.5% accuracy loss, and lightweight channel estimation achieving classical-method performance within 453.6 KB memory. Each dataset is characterized for preprocessing complexity, edge deployment feasibility, and alignment with 6G paradigms including integrated sensing and communication (ISAC), non-terrestrial networks (NTN), and mmWave operation. Through this systematic evaluation, we uncover critical gaps including limited support for mmWave and THz bands, indoor-centric bias, and inadequate scaling to high-density scenarios. We provide recommendations for dataset design and open benchmarking to enable reproducible TinyML development for 6G edge intelligence.